Re: [PATCH v13 00/10] powerpc: Switch to CONFIG_THREAD_INFO_IN_TASK

From: Christophe Leroy <christophe.leroy@c-s.fr>
To: Michael Ellerman <mpe@ellerman.id.au>,
	Benjamin Herrenschmidt <benh@kernel.crashing.org>,
	Paul Mackerras <paulus@samba.org>,
	Nicholas Piggin <npiggin@gmail.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>,
	linuxppc-dev@lists.ozlabs.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v13 00/10] powerpc: Switch to CONFIG_THREAD_INFO_IN_TASK
Date: Thu, 24 Jan 2019 09:43:47 +0000	[thread overview]
Message-ID: <9b4dab64-6a6b-a599-4676-bf891473ada7@c-s.fr> (raw)
In-Reply-To: <87y37ax3ru.fsf@concordia.ellerman.id.au>

On 01/24/2019 01:06 AM, Michael Ellerman wrote:
> Christophe Leroy <christophe.leroy@c-s.fr> writes:
>> Le 12/01/2019 à 10:55, Christophe Leroy a écrit :
>>> The purpose of this serie is to activate CONFIG_THREAD_INFO_IN_TASK which
>>> moves the thread_info into task_struct.
>>>
>>> Moving thread_info into task_struct has the following advantages:
>>> - It protects thread_info from corruption in the case of stack
>>> overflows.
>>> - Its address is harder to determine if stack addresses are
>>> leaked, making a number of attacks more difficult.
>>
>> I ran null_syscall and context_switch benchmark selftests and the result
>> is surprising. There is slight degradation in context_switch and a
>> significant one on null_syscall:
>>
>> Without the serie:
>>
>> ~# chrt -f 98 ./context_switch --no-altivec --no-vector --no-fp
>> 55542
>> 55562
>> 55564
>> 55562
>> 55568
>> ...
>>
>> ~# ./null_syscall
>>      2546.71 ns     336.17 cycles
>>
>>
>> With the serie:
>>
>> ~# chrt -f 98 ./context_switch --no-altivec --no-vector --no-fp
>> 55138
>> 55142
>> 55152
>> 55144
>> 55142
>>
>> ~# ./null_syscall
>>      3479.54 ns     459.30 cycles
>>
>> So 0,8% less context switches per second and 37% more time for one syscall ?
>>
>> Any idea ?
> 
> What platform is that on?

It is on the 8xx

> 
> On 64-bit we have to turn one mtmsrd into two and that's obviously a
> slow down. But I don't see that you've done anything similar in 32-bit
> code.
> 
> I assume it's patch 8 that causes the slow down?

I have not digged into it yet, but why patch 8 ?

I run null_syscall with perf, and I get the following. Can we conclude 
on something with that ?

Without the serie:

# Overhead       Samples  Command       Shared Object      Symbol 

# ........  ............  ............  ................. 
........................................
#
     32.95%         46375  null_syscall  [kernel.kallsyms]  [k] DoSyscall
     23.64%         33275  null_syscall  [kernel.kallsyms]  [k] 
__task_pid_nr_ns
     15.47%         21778  null_syscall  libc-2.23.so       [.] 
__GI___getppid
      8.92%         12556  null_syscall  [kernel.kallsyms]  [k] 
__rcu_read_unlock
      5.69%          8014  null_syscall  [kernel.kallsyms]  [k] sys_getppid
      4.01%          5643  null_syscall  [kernel.kallsyms]  [k] 
__rcu_read_lock
      3.67%          5166  null_syscall  [kernel.kallsyms]  [k] 
syscall_dotrace_cont
      2.52%          3542  null_syscall  null_syscall       [.] main

With the serie:

     30.04%         56337  null_syscall  [kernel.kallsyms]  [k] DoSyscall
     13.89%         26060  null_syscall  [kernel.kallsyms]  [k] 
__rcu_read_unlock
     13.36%         25062  null_syscall  libc-2.23.so       [.] 
__GI___getppid
     12.73%         23872  null_syscall  [kernel.kallsyms]  [k] 
__task_pid_nr_ns
     11.21%         21033  null_syscall  [kernel.kallsyms]  [k] sys_getppid
      8.24%         15457  null_syscall  [kernel.kallsyms]  [k] 
syscall_dotrace_cont
      4.38%          8217  null_syscall  [kernel.kallsyms]  [k] 
ret_from_syscall
      2.54%          4773  null_syscall  null_syscall       [.] main

Christophe